Determining the impact of putative loss-of-function variants in protein-coding genes

Suganthi Balasubramanian, Yao Fu, Mayur Pawashe, Patrick D. McGillivray, Mike Jin, Jeremy Liu, Konrad J. Karczewski, Daniel G. MacArthur, Mark GersteinbioRxiv 2017 · DOI

TL;DR

This paper introduces ALoFT, a novel computational tool that annotates and predicts the disease-causing potential of loss-of-function (LoF) variants. ALoFT can distinguish between variants that cause disease in heterozygotes versus homozygotes, and its application reveals that most individuals carry at least two heterozygous LoF variants that could be detrimental if homozygous.

What Problem Does This Paper Address?

Loss-of-function (LoF) variants, which disrupt the normal function of genes, are of great interest due to their significant clinical impact and surprisingly high prevalence in healthy individuals. However, accurately predicting the pathogenicity of these variants is challenging. Existing methods often struggle to differentiate between variants that cause disease only when inherited in two copies (homozygous) versus those that can cause disease even with a single copy (heterozygous). Furthermore, understanding the full spectrum of LoF variants in an individual's genome and their potential contribution to disease requires robust annotation and prediction tools.

Methods

The core of this work is the development of ALoFT (Annotation of Loss-of-Function Transcripts), a pipeline designed to annotate and predict the pathogenicity of LoF variants. ALoFT integrates a comprehensive set of features, including:

  • Functional annotations: Information about protein domains, post-translational modifications, and predicted nonsense-mediated decay (NMD) susceptibility.
  • Evolutionary conservation: GERP scores and dN/dS values to assess how conserved a variant's position is across species.
  • Biological networks: Proximity to known disease genes and network centrality measures.
  • Annotation quality checks: Features to identify potential mismapping and annotation errors, which are known to be enriched in LoF variant calls.

Using these annotated features, ALoFT employs a random forest prediction model to classify LoF variants into three categories: benign, recessive disease-causing (homozygous), and dominant disease-causing (heterozygous). The model was trained on curated datasets of known benign, dominant, and recessive LoF variants.

Datasets

  • Training Data:
    • Homozygous premature stop variants from healthy individuals in the 1000 Genomes Project (1KG).
  • Homozygous premature stop mutations from HGMD associated with recessive diseases.
  • Heterozygous premature stop variants in haplo-insufficient genes from HGMD associated with dominant diseases.
  • Evaluation Data:
    • ClinVar database for Mendelian disease-causing premature stop mutations.
  • De novo pLoF mutations from studies of autism-affected families.
  • Somatic pLoF variants from over 6,500 cancer exomes.
  • Premature stop variants from the 1000 Genomes Project (1KG), ESP6500, and ExAC datasets for analysis in healthy populations and last exons.
  • HGMD database for comparison with known disease-causing variants.

Key Insights and Ideas

  • Ubiquitous LoF burden: Every individual likely carries at least two heterozygous LoF variants that could cause disease if present in homozygous form, highlighting a significant genetic burden in the general population.
  • Distinguishing inheritance modes is crucial: ALoFT's ability to differentiate between dominant and recessive LoF variants is critical for accurate pathogenicity prediction and understanding disease risk.
  • Somatic LoFs in cancer are enriched in driver genes: Pathogenic somatic LoF variants are disproportionately found in genes known to drive cancer development, suggesting their early role in tumorigenesis.
  • Last exon variants are not always benign: While often assumed to escape NMD, ALoFT can distinguish between benign and disease-causing LoF variants in the last coding exon, suggesting a more nuanced understanding is needed.
  • ALoFT's broad applicability: The tool demonstrates utility across diverse applications, from Mendelian diseases and autism to cancer and personal genome interpretation.

Key Results

  • ALoFT achieves a high multiclass test AUC of 0.97 with 10-fold cross-validation.
  • Precisions for the three classes are: Dominant=0.86, Recessive=0.86, Benign=0.96.
  • When applied to de novo pLoF variants in autism, ALoFT found a significantly higher proportion of dominant disease-causing variants in patients compared to siblings.
  • In cancer exomes, ALoFT-predicted deleterious somatic LoF variants were significantly enriched in known cancer driver genes.
  • ALoFT correctly distinguished between dominant and recessive disease-causing variants from ClinVar, outperforming GERP and CADD scores in this regard.
  • Approximately 12.4% of premature stop variants in healthy individuals may affect minor isoforms, potentially explaining why they don't always manifest disease.

Limitations and Future Work

  • Gain-of-function mutations: The current model focuses on LoF effects and does not systematically model gain-of-function mutations, which can also arise from truncating variants.
  • Experimental validation: The authors suggest that further work is needed to correlate ALoFT predictions with experimental assays of protein loss of function.
  • Large clinical cohorts: Studying the phenotypic impact of heterozygous and homozygous LoF variants in large clinical cohorts is proposed as a future direction.
  • Integration with phenotypic data: Integrating benign LoF variants with phenotypic information is suggested for identifying protective variants and potential drug targets.

Summary updated

Abstract

Abstract Variants predicted to result in the loss of function (LoF) of human genes have attracted interest because of their clinical impact and surprising prevalence in healthy individuals. Here, we present ALoFT (Annotation of Loss-of-Function Transcripts), a method to annotate and predict the disease-causing potential of LoF variants. Using data from Mendelian disease-gene discovery projects, we show that ALoFT can distinguish between LoF variants deleterious as heterozygotes and those causing disease only in the homozygous state. Investigation of variants discovered in healthy populations suggests that each individual carries at least two heterozygous premature stop alleles that could potentially lead to disease if present as homozygotes. When applied to de novo pLoF variants in autism-affected families, ALoFT distinguishes between deleterious variants in patients and benign variants in unaffected siblings. Finally, analysis of somatic variants in > 6,500 cancer exomes shows that pLoF variants predicted to be deleterious by ALoFT are enriched in known driver genes.
BibTeX
@misc{balasubramanian2017determining,
  author = {Suganthi Balasubramanian and Yao Fu and Mayur Pawashe and Patrick D. McGillivray and Mike Jin and Jeremy Liu and Konrad J. Karczewski and Daniel G. MacArthur and Mark Gerstein},
  title = {{Determining the impact of putative loss-of-function variants in protein-coding genes}},
  year = {2017},
  doi = {10.1101/106468},
  howpublished = {bioRxiv},
}

Loading related content...

Figure